This has been requested so I'm putting it here for those of you that want to play with the performance of the new DLA algo with data since 2017 until Jan 24 2018. Here is the link for the .csv file: https://storage.googleapis.com/gamma-trader-positions/amb/infer_ndla_2017.csv
You can run your own statistical analysis with it. The high level take away from that data is this:
Precision: 71.19% (This is the amount of signals >=0.5 that actually worked divided by the signals generated)
Recall: 64.6% (This is the amount of signals >=0.5 divided by the total amount of 1% moves that actually happened in the market).
As you can see the NDLA improves the recall vs the old one which means it has less false negatives signals. Also the precision is improved a bit (which improves our trading edge too).